How much time does AI automation really save?
It varies, and no study tells you what your week will look like. A St. Louis Fed survey, based on what workers reported, put time saved at 1.6% of all work hours across all workers, including people who do not use generative AI. An NBER study of customer support agents found 14% more issues resolved per hour on average. The reliable answer comes from measuring your own tasks before and after.
What a St. Louis Fed survey measured
A November 2025 post from the St. Louis Fed by Alexander Bick, Adam Blandin and David Deming draws on its quarterly Real-Time Population Survey of U.S. adults ages 18 to 64. Users reported how many extra hours they would have needed to get the same work done without generative AI. Across all workers, including those who do not use it, the post puts reported time savings at 1.6% of all work hours.
These are workers' own estimates of a situation that did not happen, not timed measurements. The authors note that saved time may be spent on lower-value tasks or on-the-job leisure, which would reduce measured productivity gains. A figure across all workers also says little about any one business.
What an NBER study measured
Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied a staggered rollout of a generative AI-based conversational assistant to 5,179 customer support agents. Productivity, measured as issues resolved per hour, rose 14% on average. Novice and low-skilled workers improved by 34%, while experienced and highly skilled workers saw minimal impact.
This looked at real work rather than a lab exercise, but it was one job. Results for customer support may not carry over to quoting, scheduling or bookkeeping, and the gain was uneven between people.
How to measure your own week
Both studies describe averages for particular groups. Treat them as a reason to measure, not as a forecast. A simple before-and-after count is usually enough.
- Pick one task, such as replying to new inquiries or writing quotes.
- For about two weeks before the change, note roughly how long each one takes and how many you finish.
- Note how many need fixing afterward, such as a wrong date or a missing detail.
- Run the new way for a similar stretch, tracking the same things. Include the time spent checking the output, fixing it and keeping the setup working.
- Compare per task, not by how the week felt. A busy season in one period and a quiet one in the other can mislead.
- Ask where the saved time went: more jobs, faster replies, or nowhere you can see. If the saving is small or the errors rise, it can be reasonable to narrow the job or switch it off.
Where to go next
The related answer on whether a task is worth automating scores a task before you start, and the related answer on whether a person should approve what AI sends to customers covers the review step, whose time belongs in your count.
Sources
- Bick, Blandin and Deming, St. Louis Fed, "The State of Generative AI Adoption in 2025" (November 13, 2025)
- Brynjolfsson, Li and Raymond, NBER Working Paper 31161, "Generative AI at Work" (April 2023, revised November 2023)
Related answers
- How do you tell if a task is worth automating?
- Should a person approve what an AI system sends to customers?
- What can a small business realistically automate with AI?
AI systems and automation
A lot of an owner's week goes to the same tasks: answering the same questions, chasing the same follow-ups, retyping the same details. We build the systems that do that work, so your people can spend their time on customers.